feat(observability): expand Sentry instrumentation for self-host reviews

[JSONbored]

Summary

Add stronger Sentry instrumentation across the self-host review engine so provider failures, queue failures, config failures, and posting failures have enough context to debug quickly.

Context

When reviews stop firing or dashboards go stale, the stack should surface actionable failure context instead of requiring log spelunking. Sentry should capture the important boundaries without leaking secrets, repo-private data, tokens, or raw review prompts.

Requirements

• Inventory current Sentry coverage across webhook intake, queue processing, review planning, provider execution, synthesis, posting, metrics, and config loading.
• Add structured breadcrumbs/tags for provider, repo, review mode, event type, queue job, failure class, and retry state.
• Scrub secrets, tokens, prompts, private diff content, and auth payloads.
• Capture known failure classes distinctly: provider quota/auth, CLI spawn failure, config validation, GitHub API failure, queue timeout, and synthesis failure.
• Document how to trace a failed review from dashboard symptoms to Sentry issue context.

Deliverables

• Sentry instrumentation map.
• Code changes adding missing spans, tags, breadcrumbs, and error classes.
• Sanitization tests or explicit scrubber coverage.
• Troubleshooting docs for common self-host review failures.

Acceptance criteria

• A failed review has enough Sentry context to identify the failing boundary without exposing sensitive data.
• Provider quota/auth failures are distinguishable from review logic failures.
• Tests prevent accidental secret or prompt leakage in Sentry payloads.